基于深度残差收缩网络的船用补水泵滚动轴承故障诊断

Fault Diagnosis of Rolling Bearings in Marine Charge Pumps Based on Deep Residual Shrinkage Network

船用补水泵是常规的船用设备,与陆上普通补水泵不同的是船用补水泵有着较高的可靠性要求,并且要求故障发生时要及时发现故障位置。为了能够提升对于补水泵的健康监测以及智能故障诊断,这篇文章提出了一种基于深度残差收缩网络的补水泵滚动轴承故障诊断模型。该模型使用的深度残差收缩网络是对于残差网络的改进,首先增加了网络深度,强化了特征提取能力,残差模块的显著特点是具有恒等映射结构,该结构能有效解决深度神经网络中的梯度消失或爆炸问题。通过软阈值和注意力机制的深度融合从而实现样本降噪功能。最后,为了验证方法的有效性,采用大量的补水泵滚动轴承振动信号进行测试,通过与其他主流网络模型的故障分类准确率对比,得出结论深度残差收缩网络对于滚动轴承的故障具有较高的分类精度。

Marine charge pump is a more conventional marine equipment,different from land-based ordinary charge pumps,marine charge pumps have higher reliability requirements,and require timely detection of fault location when failure occurs.In order to improve the health monitoring and intelligent fault diagnosis of the charge pump,this paper proposes a fault diagnosis model for the rolling bearing of the charge pump based on deep residual shrinkage network.The deep residual systolic network used in this model is an improvement of the residual network,which firstly increases the network depth and thus strengthens the feature extraction ability.The significant feature of the residual module is that it has a constant mapping structure,which can effectively solve the gradient disappearance or explosion problem in the deep neural network.A deep fusion of soft thresholding and attention mechanism is used to achieve the sample noise reduction function.Finally,in order to verify the effectiveness of the method,a large number of rolling bearing vibration signals of the charge pump are used for testing,and by comparing the fault classification accuracy with other mainstream network models,it is concluded that the deep residual contraction network has a high classification accuracy for rolling bearing faults.